BSC-LT/speech-salamandra-es-en
Partial fine-tune (fewer layers) vs BSC-LT/salamandra-7b-instruct. BSC-LT/salamandra-7b-instruct matches part of BSC-LT/speech-salamandra-es-en (291 of its 1095 tensors; the rest of BSC-LT/speech-salamandra-es-en has no counterpart). On the shared part: Partial change: 87% of the weights changed (attention, lm_head, mlp, norms), in 32 of 32 layers.
Tensors changed: Embeddings 0/1, Norms 65/65, MLP 96/96, Attention 128/128, lm_head 1/1